ConvNext-Base: Optimized for Qualcomm Devices

ConvNextBase is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of ConvNext-Base found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
ONNX	float	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
ONNX	w8a16	Universal	QAIRT 2.37, ONNX Runtime 1.23.0	Download
QNN_DLC	float	Universal	QAIRT 2.42	Download
QNN_DLC	w8a16	Universal	QAIRT 2.42	Download
TFLITE	float	Universal	QAIRT 2.42, TFLite 2.17.0	Download

For more device-specific assets and performance metrics, visit ConvNext-Base on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for ConvNext-Base on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_classification

Model Stats:

• Model checkpoint: Imagenet
• Input resolution: 224x224
• Number of parameters: 88.6M
• Model size (float): 338 MB
• Model size (w8a16): 88.7 MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
ConvNext-Base	ONNX	float	Snapdragon® X Elite	7.488 ms	176 - 176 MB	NPU
ConvNext-Base	ONNX	float	Snapdragon® 8 Gen 3 Mobile	5.436 ms	0 - 394 MB	NPU
ConvNext-Base	ONNX	float	Qualcomm® QCS8550 (Proxy)	7.317 ms	0 - 638 MB	NPU
ConvNext-Base	ONNX	float	Qualcomm® QCS9075	11.598 ms	0 - 4 MB	NPU
ConvNext-Base	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	4.246 ms	0 - 329 MB	NPU
ConvNext-Base	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	3.352 ms	0 - 332 MB	NPU
ConvNext-Base	ONNX	w8a16	Snapdragon® X Elite	219.265 ms	137 - 137 MB	NPU
ConvNext-Base	ONNX	w8a16	Qualcomm® QCS6490	1158.253 ms	41 - 86 MB	CPU
ConvNext-Base	ONNX	w8a16	Qualcomm® QCS9075	317.703 ms	93 - 96 MB	NPU
ConvNext-Base	ONNX	w8a16	Qualcomm® QCM6690	737.008 ms	34 - 46 MB	CPU
ConvNext-Base	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	268.327 ms	78 - 229 MB	NPU
ConvNext-Base	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	691.251 ms	35 - 49 MB	CPU
ConvNext-Base	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	237.073 ms	89 - 240 MB	NPU
ConvNext-Base	QNN_DLC	float	Snapdragon® X Elite	8.589 ms	1 - 1 MB	NPU
ConvNext-Base	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	6.113 ms	0 - 350 MB	NPU
ConvNext-Base	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	42.453 ms	1 - 279 MB	NPU
ConvNext-Base	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	8.213 ms	0 - 33 MB	NPU
ConvNext-Base	QNN_DLC	float	Qualcomm® QCS9075	12.381 ms	1 - 3 MB	NPU
ConvNext-Base	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	20.603 ms	0 - 337 MB	NPU
ConvNext-Base	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	4.689 ms	1 - 281 MB	NPU
ConvNext-Base	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	3.534 ms	1 - 283 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Snapdragon® X Elite	6.26 ms	0 - 0 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	4.106 ms	0 - 248 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Qualcomm® QCS6490	23.818 ms	0 - 2 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	14.472 ms	0 - 199 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	5.888 ms	0 - 2 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Qualcomm® QCS9075	6.122 ms	0 - 2 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Qualcomm® QCM6690	71.461 ms	0 - 395 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	9.182 ms	0 - 246 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	3.31 ms	0 - 190 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	7.715 ms	0 - 247 MB	NPU
ConvNext-Base	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	2.559 ms	0 - 201 MB	NPU
ConvNext-Base	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	5.533 ms	0 - 345 MB	NPU
ConvNext-Base	TFLITE	float	Qualcomm® QCS8275 (Proxy)	41.241 ms	0 - 274 MB	NPU
ConvNext-Base	TFLITE	float	Qualcomm® QCS8550 (Proxy)	7.334 ms	0 - 3 MB	NPU
ConvNext-Base	TFLITE	float	Qualcomm® QCS9075	11.149 ms	0 - 177 MB	NPU
ConvNext-Base	TFLITE	float	Qualcomm® QCS8450 (Proxy)	19.73 ms	0 - 330 MB	NPU
ConvNext-Base	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	4.167 ms	0 - 277 MB	NPU
ConvNext-Base	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	3.174 ms	0 - 279 MB	NPU

License

• The license for the original implementation of ConvNext-Base can be found here.

References

• A ConvNet for the 2020s
• Source Model Implementation

Community

• Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
• For questions or feedback please reach out to us.